EUVIMEDEUVIMED entdecken
Crossref

MacPrompt: Maraconic-Guided Jailbreak Against Text-to-Image Models

Xi Ye, Yiwen Liu, Lina Wang, Run Wang, Geying Yang, Yufei Hou, Jiayi Yu

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (42) · 36003-36011

DOI: 10.1609/aaai.v40i42.40916 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Xi Ye und 6 weiteren Autor:innen. Die Arbeit wurde bisher 1-mal zitiert (Crossref).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Text-to-image (T2I) models have raised increasing safety concerns due to their capacity to generate NSFW and other banned objects. To mitigate these risks, safety filters and concept removal techniques have been introduced to block inappropriate prompts or erase sensitive concepts from the models. However, all the existing defense methods are not well prepared to handle diverse adversarial prompts. In this work, we introduce MacPrompt, a novel black-box and cross-lingual attack that reveals previously overlooked vulnerabilities in T2I safety mechanisms. Unlike existing attacks that rely on synonym substitution or prompt obfuscation, MacPrompt constructs macaronic adversarial prompts by performing cross-lingual character-level recombination of harmful terms, enabling fine-grained control over both semantics and appearance. By leveraging this design, MacPrompt crafts prompts with high semantic similarity to the original harmful inputs (up to 0.96) while bypassing major safety filters (up to 100%). More critically, it achieves attack success rates as high as 92% for sex-related content and 90\% for violence, effectively breaking even state-of-the-art concept removal defenses. These results underscore the pressing need to reassess the robustness of existing T2I safety mechanisms against linguistically diverse and fine-grained adversarial strategies. Warning: This paper includes sensitive examples (e.g., adult, violent, or illegal content). Unsafe images are masked but may still be disturbing.

Zitieren

APA

Ye, X., Liu, Y., Wang, L., Wang, R., Yang, G., Hou, Y., & Yu, J. (2026). MacPrompt: Maraconic-Guided Jailbreak Against Text-to-Image Models. Proceedings of the AAAI Conference on Artificial Intelligence, 40(42), 36003-36011. https://doi.org/10.1609/aaai.v40i42.40916

Vancouver

Ye X, Liu Y, Wang L, Wang R, Yang G, Hou Y, et al. MacPrompt: Maraconic-Guided Jailbreak Against Text-to-Image Models. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(42):36003-36011. doi:10.1609/aaai.v40i42.40916

BibTeX
@article{Ye2026,
  author = {Ye, Xi and Liu, Yiwen and Wang, Lina and Wang, Run and Yang, Geying and Hou, Yufei and Yu, Jiayi},
  title = {MacPrompt: Maraconic-Guided Jailbreak Against Text-to-Image Models},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {42},
  pages = {36003--36011},
  doi = {10.1609/aaai.v40i42.40916}
}
Kostenlos starten →